Bayesian Methods in Animal Breeding with Different Prior Specification Using Gibbs Sampling

Abstract

An application of Gibbs sampling in animal breeding is carried out using two different types of prior specification. The first specification has full conjugacy and the second one has non-conjugacy, which causes computational difficulties. This paper demonstrates how the Gibbs sampling procedure, making use of an adaptive rejection sampling algorithm, deals with non-conjugacy in animal breeding using a balanced one-way classification with a random sire effects model and compares the results of two different specifications. It is also shown how Bayesian methods solve the problem of obtaining a negative estimated variance.